JP2019531065A - Method, system and computer program product for determining the presence of microorganisms and identifying said microorganisms - Google Patents

Abstract

Method and system for determining the presence of at least one identified microorganism in a Petri dish comprising one or more colonies of microorganism and medium, and when implemented, the method of the present invention can be performed Computer program product to make. [Selection] Figure 5

Description

  The present invention relates to the field of microbiological analysis for analyzing a biological sample in a Petri dish and containing microorganisms, and more specifically, the microorganisms such as bacteria in a Petri dish. The present invention relates to a method, a system, and a computer program for determining presence and identifying the microorganism.

  In the field of microbiology, analysis is often performed on the growth of biological samples containing, for example, microorganisms, cell extracts or intracellular extracts in order to determine the presence of specific bacteria and to identify such bacteria. Are known. Thus, the corresponding disease can be diagnosed.

  There are several solutions in the prior art for identifying bacteria. The solution is based on a method that includes a supervised learning algorithm that uses criteria called descriptors. Descriptors relate to bacterial color, morphology and structure.

  Such a method also operates with steps for determining the location of isolated bacterial colonies. The location may be provided by the user in a semi-automated form or by applying a specific segmentation algorithm in an automated fashion. However, both of these methods of providing position are inaccurate and generate several errors that lead to incorrect position determination.

  US Patent Application Publication No. 2015/0087017 discloses a method for automatic classification of bacterial colonies. The method is based on a classification step of bacterial colonies in a Petri dish, which is based on criteria such as the color, shape and contour of each bacterial colony.

  However, such an automated method is not applicable to the analysis of the contents of a petri dish. In fact, the environment around the bacterial colonies in the Petri dish must also be analyzed to identify the specific biological phenomena associated with the bacterial colonies and their interaction with the medium. Such biochemical reactions are used by biologists to identify various microorganisms and make it possible to make a diagnostic hypothesis or reject it.

For example, with respect to analyzing the growth of microorganisms in a Petri dish containing blood agar, the prior art method is a hemolytic phenomenon that determines the level at which blood cells in the agar are degraded to characterize a particular microorganism. Includes steps to identify properties. Hemolysis is associated with the ability of certain microorganisms to produce proteins that can destroy agar blood cells. There are three types of hemolysis phenomena named alpha, beta and gamma as follows:
-Alpha hemolysis causes a greenish discoloration surrounding bacterial colonies growing on blood agar. This type of hemolysis represents a partial degradation of hemoglobin in red blood cells of blood agar.
-The beta hemolysis phenomenon represents the total degradation of hemoglobin in erythrocytes in the vicinity of bacterial colonies.
-Gamma hemolysis corresponds to the absence of any degradation in the area surrounding the bacterial colonies.

  In other situations, the interaction between the bacteria and the medium produces visible artifacts surrounding the colony, such as the appearance of a particular color. These color changes are due to a chemical reaction between the culture medium and the metabolites produced by the bacteria, which can be interpreted as a positive or negative reaction depending on the nature of the microorganism itself. Today, such a step is provided by a user who has to look at the contents of a Petri dish to determine the nature of the hemolysis phenomenon or the color change around the colony. However, such human analysis can be error prone. In addition, in situations where the number of petri dishes to be analyzed is increasing, the identification step is a time consuming process.

  Thus, there is a need to improve the prior art methods to enable automatic identification of biological elements in petri dishes in an efficient manner.

According to one aspect of the invention, a method for determining the presence of at least one identified microorganism (eg, at least one identified bacteria) in a Petri dish comprising one or more colonies of microorganism and a culture medium. A method wherein the medium is adapted to allow one or more colonies of the microorganism and, if present, the at least one identified microorganism to grow under suitable growth conditions In
-Obtaining at least one initial image of the Petri dish, wherein the first initial image comprises one or more visible pixels, each pixel being associated with a pixel value;
-Obtaining a first processed image of the Petri dish by applying a first process to at least one initial image, wherein the visible pixels of the first processed image are one or more colonies of microorganisms; Steps associated only with pixels associated with
-Obtaining a plurality of second processed images of the Petri dish by applying a second processing to the first processed image, wherein the visible pixels of the second processed image are one or more of the microorganisms; Relating only to pixels associated with a colony of microorganisms and surrounding zones around said one or more colonies of microorganisms;
-Obtaining a plurality of third processed images of the Petri dish by calculating a difference between at least one initial image and a plurality of second processed images, the plurality of third processing The completed image includes visible pixels associated only with pixels associated with the culture medium;
-Obtaining a plurality of fourth processed images of the Petri dish by calculating a difference between the plurality of second processed images and the first processed image, the plurality of fourth processed images The visible pixels of the processed image are related only to the pixels associated with the surrounding zone;
By calculating a difference value between an average pixel value of a plurality of third processed images and an average pixel value of a plurality of fourth processed images for at least each color channel of a red green blue (RGB) color channel; Determining features associated with the plurality of third and fourth processed images;
-Determining the value of the indicator of the presence of at least one identified microorganism in the surrounding zone in the Petri dish by classifying the determined features;
-Comparing the value of the indicator with a threshold;
-A method is provided comprising determining the presence of at least one identified microorganism in the surrounding zone in the Petri dish, depending on the result of the comparison.

In the context of this application, the term “average pixel value” refers to:
-Arithmetic mean of pixel values, or-median of pixel values.

  According to a preferred embodiment of the present invention, the feature amounts associated with the plurality of third processed images and the fourth processed image are the central pixel value of the plurality of third processed images and the plurality of fourth processed images. The difference value between the processed image and the central pixel value of the processed image is determined by calculating for at least each color channel of the red green blue (RGB) color channel.

  Preferably, the first process is a segmentation process.

  The method according to the invention comprises the automatic identification of specific zones around microbial colonies and the presence of bacteria, such as beta-hemolytic or proteus, respectively associated with biological phenomena such as hemolysis or Proteus infection Provides automatic analysis of petri dish contents based on automatic determination of Thus, the method avoids the operations of the visual identification step and / or the visual determination step.

  The method according to the invention comprises a step for identifying a specific zone around a colony of microorganisms in a petri dish, said specific zone comprising a bacterial colony or a bacterial cluster of colonies.

According to a further aspect of the invention, for determining the presence of at least one identified microorganism (eg, at least one identified bacteria) in a Petri dish comprising one or more colonies of microorganism and a medium. A system wherein the medium is adapted to allow one or more colonies of the microorganism and, if present, the at least one identified microorganism to grow under suitable growth conditions. , A system comprising an imaging system for acquiring at least one initial image of a Petri dish, and a processing system, wherein the processing system comprises:
-A first processing unit for obtaining a first processed image of at least one initial image of the petri dish, wherein the visible pixels of the first processed image are associated with one or more colonies of the microorganism; A first processing unit relating only to the selected pixel;
-A second processing unit for obtaining a plurality of second processed images of the first processed image of the petri dish, wherein the visible pixels of the second processed image are one or more colonies of the microorganism; And a second processing unit relating only to pixels associated with the surrounding zone around the one or more colonies of microorganisms;
-Obtaining a plurality of third processed images of the Petri dish by calculating a difference between the second initial image and the second processed image; and the second processed image and the first processed image A calculation unit for obtaining a plurality of fourth processed images of a Petri dish by calculating the difference between the processed images, wherein the third processed image is only for the pixels associated with the medium. A computing unit that is associated only with pixels whose visible pixels of the fourth processed image are associated with the surrounding zone;
A plurality of third processes by calculating, for at least each color channel of the RGB color channel, a difference value between the average pixel value of the third processed image and the average pixel value of the fourth processed image; A feature extraction unit for determining a feature quantity associated with the completed image and the fourth processed image;
Determining the value of the indicator of the presence of at least one identified microorganism in the surrounding zone in the Petri dish and comparing the value of the indicator with a threshold; and depending on the result of the comparison, the surrounding zone in the Petri dish A system is provided comprising an analysis unit for determining the presence of at least one identified microorganism within.

  As indicated above, according to a preferred embodiment of the present invention, the feature quantities associated with the plurality of third processed images and the fourth processed image are the center of the plurality of third processed images. A difference value between the pixel value and the center pixel value of the plurality of fourth processed images is determined by calculating for at least each color channel of the red green blue (RGB) color channel.

  According to a further aspect of the present invention there is provided a computer program product comprising instructions that, when executed, cause a programmable data processing device to perform the steps of the method according to the present invention.

  Reference will now be made to the accompanying drawings as an example.

1 represents a schematic diagram of a system according to an embodiment of the present invention. Fig. 3 represents a wheel-form controller with circular holes, white background circles, and black background circles, according to one embodiment of the invention. 1 is a schematic diagram of an imaging system showing different types of illumination beams irradiated on a petri dish according to an embodiment of the present invention. FIG. 2 shows a flowchart of a method for determining the presence of bacteria according to one embodiment of the present invention. 2 shows a flowchart of a method for determining the presence of hemolytic bacteria according to one embodiment of the present invention. Fig. 4 represents an original image of a Petri dish containing blood agar and hemolytic bacteria according to an embodiment of the invention, said original image being related to backlight illumination. FIGS. 7a, 7b and 7c show the original red, green and blue color channel images, respectively, with a central bottom view, according to one embodiment of the present invention. Fig. 3 shows a binary image called GrowthMask of the original image, according to one embodiment of the present invention. FIG. 9a shows an image associated with the HaloMask of the original image of FIG. 6, according to one embodiment of the present invention. FIG. 9b shows the Culture Medium Mask of the original image of FIG. 6, according to one embodiment of the invention. FIGS. 10a and 10b show Halo Mask and Culture Medium Mask, respectively, for a 200 × 200 pixel pixel patch with an extension parameter of 50 pixels, according to one embodiment of the present invention. FIG. 11a shows a HaloMask pixel patch with an extension parameter of 10 pixels, according to one embodiment of the present invention. FIG. 11b shows a Culture Medium Mask pixel patch with an extension parameter of 10 pixels according to one embodiment of the present invention. FIG. 12a shows a HaloMask pixel patch with an extension parameter of 50 pixels, according to one embodiment of the present invention. FIG. 12b shows a Culture Medium Mask pixel patch with an extension parameter of 50 pixels, according to one embodiment of the present invention. FIG. 13a shows a HaloMask pixel patch with an extension parameter of 60 pixels, according to one embodiment of the present invention. FIG. 13b shows a Culture Medium Mask pixel patch with an extension parameter of 60 pixels, according to one embodiment of the present invention. FIG. 4 shows the distribution of features at the patch level in a Petri dish containing beta hemolytic bacteria, alpha hemolytic bacteria, and gamma hemolytic bacteria, according to one embodiment of the present invention. FIG. 4 represents a graph showing the distribution of pixel patches for color difference values between HaloMask and CultureMediumMask on a blue channel, according to one embodiment of the present invention. 2 shows a flowchart of a method for determining the presence of Proteus according to one embodiment of the present invention. FIG. 16a shows an original image of a Petri dish containing CPS medium and Proteus according to one embodiment of the present invention. FIG. 16b shows a binary image called Binary Growth Mask of the original image of FIG. 16a, according to one embodiment of the present invention. FIG. 17a shows a colored GrowthMask of the original image of FIG. 16a, according to one embodiment of the present invention. FIG. 17b shows the colored HaloMask of the original image of FIG. 16a, according to one embodiment of the present invention. FIG. 17c shows the colored CutureMask of the original image of FIG. 16a, according to one embodiment of the present invention. FIG. 18a shows the profile of HaloMask overlaid with the bottom circular view of a Petri dish containing CPS medium and class 4 bacteria, with an expansion parameter of 70 pixels, according to one embodiment of the present invention. FIG. 18b shows the profile of HaloMask overlaid with the bottom circular view of a Petri dish containing CPS medium and class 4 bacteria, with an expansion parameter of 10 pixels, according to one embodiment of the present invention. The graph showing the distribution of the proteus presence probability according to one embodiment of the present invention is shown. FIG. 4 shows an original image of a Petri dish containing an opaque medium, such as an opaque CPS containing Proteus, according to one embodiment of the present invention. FIG. 21 shows an image of the Petri dish of FIG. 20 related to the left bottom annular view, according to one embodiment of the present invention. FIG. 21 shows an image of the Petri dish of FIG. 20 related to the upper annular view, according to one embodiment of the present invention. Distribution of proteus presence probability p (Halo / X) using CART algorithm and SVM algorithm and distribution of final probability of Proteus presence to Petri dish containing translucent medium according to one embodiment of the present invention The graph which shows is shown. Distribution of the probability of the presence of Proteus using the CART algorithm and the SVM algorithm, p (Halo / X), and the distribution of the final probability of the presence of Proteus on a Petri dish containing an opaque medium, according to one embodiment of the invention The graph which shows is shown.

  The following description discloses the invention in a sufficiently clear and complete manner based on specific embodiments. However, this description should not be construed as limiting the scope of protection to the specific embodiments and examples described below.

  In the following description, the term “Petri dish” defines an assembly of a Petri dish and a lid covering the Petri dish. The Petri dish can include a serigraphy shown on the Petri dish or on the lid.

  In the following description, the term “pixel” relates to a position in the image, and the term “pixel value” refers to a detected value of the corresponding capture intensity of the pixel in the image.

  In the following description, the term “dark pixel” relates to a pixel having a pixel value of zero.

  In the following description, the term “visible pixel” relates to a pixel having a value other than 0, also known as a bright pixel.

  The described calculation operation relates to an operation associated with the pixel value of the associated image.

  In the following description, the representation of the surface of the Petri dish is performed in orthonormal space x, y, z, where z is a function of x and y, such as z = f (x, y).

  In the following description, the pixels associated with the Petri dish define a zone in the image called the DishMask.

  In the following description, the pixels associated with the media define a media zone in the image called Culture Medium Mask.

  Similarly, pixels located around the microbial colony define a surrounding or halo zone in the image called HaloMask.

  In the following description, the term “original image” relates to an image of a Petri dish that has been acquired using an image capture device.

  In the following description, the term “initial image” refers to an image of a Petri dish, where the initial image may relate to an original image or an image that has been acquired in an image capture device and then processed in an imaging system.

  FIG. 1 shows an example of a system 100 according to the present invention. The system 100 includes a sample container bank 102, an automatic line drawing device 104, a smart incubator system 106, a processing system 108 and an analysis unit 110.

  The sample container bank 102 manually or automatically creates a sample container within which a biological sample can be grown and analyzed. The sample container is typically a petri dish, but other containers may also be used. Accordingly, references herein to petri dishes are not intended to be limiting.

  The sample container bank 102 adds the appropriate medium to the petri dish so that the biological sample can be grown.

  The automatic streak device 104 applies the biological sample to a petri dish and then distributes the biological sample in a known manner. For example, a sample is applied to a petri dish using a comb having a length approximately equal to the radius of the petri dish. A comb is applied and then turned to spread the biological sample over the entire surface of the Petri dish. An example of a suitable automatic line drawing device 104 is marketed by the present applicant under the trade name PREVI® Isola.

  Once the biological sample has been distributed across the medium in the Petri dish, the Petri dish is transferred to the next processing stage, ie, the smart incubator system 106, manually by an operator or by a conveyor belt or other automated system. Is done.

  The smart incubator system 106 includes an incubator 112 and an imaging system 114. The petri dish is introduced into the smart incubator system 106 and incubated at a predetermined temperature for a predetermined time. This allows the biological sample to grow and create a large number of microbial colonies across the surface of the medium in the Petri dish. After the petri dish is incubated as required, the petri dish is transferred to the imaging system 114. The imaging system 114 is a system for creating images of colonies and cultures generated in the system 100 as a whole. Details of the imaging system are further described below.

  The image is used in the first stage of analysis of the biological sample. This stage can identify colonies and other aspects of the biological sample to support and facilitate further operation and function of the overall system 100.

  After the image of the Petri dish is generated, the Petri dish is then transferred to the next processing stage, the processing system 108. This can be done automatically by a conveyor belt or other automated system or by an operator.

  The processing system 108 can take a variety of different forms depending on the biological sample analysis required. For example, specific colonies can be extracted based on the images for further analysis or processing. Many other treatments can be applied to the Petri dish at this time. If necessary, the Petri dish can be returned to the smart incubator system 106 and / or returned to the imaging system 114 for further growth.

  After all necessary processing and imaging is complete, the Petri dish can be transferred to the analysis unit 110 manually or by an automated process.

  As previously mentioned, the smart incubator system 106 includes an imaging system. The imaging system 114 includes a base unit. The base unit includes an optical system and a control circuit for generating red, green and blue backlight illumination. The base unit may comprise a wheel-type controller taking the three positions shown in FIG. These positions correspond to different illuminations which are “no background”, “white background” and “black background”. The “no background” position is associated with the circular hole 150 in the wheel. The “white background” position is associated with the white background circle 160 of the wheel. The “black background” position is relative to the black background circle 170 of the wheel. No background is used for the backlight, while white and black are used for any other type of illumination, depending on the nature of the sample.

  There is a sample holding unit above the base unit. The sample holding unit may be provided with a drawer that can be slid in and out and provided with a recess adapted to support the Petri dish. In addition, the sample holding unit comprises four red, green and blue horizontal illumination sources. The four illumination sources are positioned linearly around the sample recess and can be controlled independently. In use, the top surface of the Petri dish is substantially aligned with the tops of the four horizontal illumination sources. A horizontal illumination source allows the Petri dish to be illuminated with a horizontal or near horizontal beam.

  It should be noted that the bottom of the sample recess is optically transmissive so that the backlight illumination can illuminate the biological sample in use. The sample holding unit further comprises the optics and controller required to operate the four horizontal illumination sources.

  The sample holding unit may comprise another orientation in which the biological sample is transferred by the conveyor belt to a position for imaging. The drawer can be replaced by a conveyor belt system having a sample holding zone that is transparent so that each can use a backlight. The conveyor belt system can move the biological sample to the appropriate location and then take the necessary images. The conveyor belt then moves the next biological sample to the imaging position and moves the first biological sample to the next processing stage. This allows images to be taken at different locations and when the biological sample is moving.

  In a further alternative, the imaging system can include a robotic arm that can place the Petri dish in the sample holder or on a conveyor belt. In addition, the robotic arm can remove the lid of the Petri dish before imaging, and then reposition the lid. This can be done by inverting the Petri dish and dropping the lid. Removing the lid ensures that the lid does not cause reflections when the biological sample is illuminated by a predetermined illumination source.

  In addition to taking in and out of the imaging zone, the sample holding unit may also include a mechanism for changing the position of the biological sample relative to the normal position. For example, the sample holding unit may allow the sample to be oriented at a specific angle with respect to a specific beam. Other movements, such as rotation of the biological sample, can also be performed by suitable mechanisms. As a result, any relative movement of the biological sample and the illumination source can be achieved by moving either the biological sample or the illumination source in the sample holding unit. The deformation is infinite.

  The imaging system 114 further comprises a first intermediate unit positioned above the sample holding unit. The first intermediate unit comprises four red, green and blue illumination sources arranged in a straight line. In use, the illumination source is adapted to produce an annular illumination on the sample recess in the sample holding unit and is independently controllable. The annular illumination can be adjusted to be incident on the sample from any suitable direction, including lateral, non-lateral, or any other suitable orientation.

  The imaging system 114 also includes a second intermediate unit. The second intermediate unit comprises four red, green and blue illumination sources arranged in a straight line. The illumination source is turned up and reflected from above the unit to produce an inverted annular illumination that illuminates the sample in the sample recess during use. Each illumination source can be controlled independently.

  The head unit of the imaging system 114 is positioned above the second intermediate unit. The head unit includes a white light illumination source that can be independently controlled. Eight illumination sources are arranged to provide vertical illumination on the sample recess in use.

  The head unit also includes an image capture device 254 such as a camera that is directed to the sample recess. Illumination by any combination of illumination sources of any of the units can be directed to the sample recess. The image capture device can then capture an image from any biological sample in the illuminated sample recess. The use of images and further processing is described in more detail below.

  The head unit may also include a control pad that is used to operate various light sources. In addition to the control circuit and optical system in each unit that controls the function, there may be an overall control system. The overall control system may comprise a computer, display unit, processing module and image enhancement algorithm, image processing, and any other process or technology.

  The overall control system can be used to control which illumination source is used for a particular application. In addition, the overall control system can apply different image enhancement techniques and image processing for different applications. Image enhancement techniques are methods and techniques that improve the image quality of an image or visualize related information for professional viewing. As will be described in more detail below, examples include integration of different images such as vertical integration or integration for hemolysis, edge illumination correction, exposure time correction, and the like. Image processing is the extraction of information from an image to provide decision support or automatic decision. This does not necessarily include image modifications, but includes automatically determining higher levels of information / interpretation. As described in more detail below, examples include dish ring detection, mark detection, growth (population, isolated colony, swarming) detection, overall determination regarding growth / non-growth, and the like.

The overall control system can be used to perform any other function and / or control action on the imaging system. These include, but are not limited to:
-Addition and removal of samples into sample recesses;
-Checking and adjusting the sample positioning in the sample recess;
-Control of the brightness level;
-Control of the red, green and blue component balance;
-Control of exposure time;
-Control of lighting combinations;
-Inspection of the system;
-Calibration of the system; and-any other appropriate control based on usage and analytical purposes.

  Each of the units forming the imaging system can be moved relative to other units. When this is done, certain optical adjustments may be necessary to ensure that the biological sample is illuminated by all illumination sources.

  The operation of the imaging system 114 will be described in more detail below with reference to FIG.

  FIG. 3 is a schematic diagram of the imaging system 114 showing various illumination sources and how these illumination sources affect the biological sample located in the imaging system 114. The sample can be illuminated by substantially horizontal beams 302, 304 located in the same plane as the plane of the Petri dish containing the biological sample. The substantially horizontal beam actually includes components orthogonal to the plane of the paper in addition to those indicated by reference numerals 302 and 304. The substantially horizontal beams 302, 304 are created by a horizontal illumination source in the sample holding unit.

  Samples can also be generated by a base unit under a Petri dish and illuminated by a backlight beam 306 that emits vertically from the bottom of the imaging system 114 toward the top of the imaging system 114.

  An annular beam or bottom annular beam 308 can also illuminate the biological sample 300 and is created by a first intermediate unit located above the Petri dish. The bottom annular beam 308 is emitted at a fixed angle toward the petri dish. An inverted or upper annular beam 310 created by a second intermediate unit positioned above the Petri dish can also illuminate the Petri dish. The inverted annular beam 310 is emitted at a determined angle toward the top of the imaging system 114 in a direction away from the Petri dish and then reflected toward the Petri dish.

  A vertical beam 312 can also illuminate the Petri dish and is generated by an illumination source in the head unit.

  The vertical beam and backlight illumination illuminate in a direction substantially orthogonal to the biological sample in the petri dish. Thus, the optical axis of each of these illumination sources is also orthogonal to the biological sample. The substantially horizontal illumination, the annular illumination and the inverted annular illumination are not orthogonal to the Petri dish. Thus, similarly, the optical axes of these light sources are non-orthogonal to the biological sample. Non-orthogonal sources provide a diverse range or alternative image to that achieved with orthogonal sources. These non-orthogonal sources provide additional and different optical features to any image created with them. This ensures improved colony isolation and detection.

  The illumination source shown in FIG. 3 and produced by a suitable unit can be any preferred type, for example, a light emitting diode (LED) operating at red, green and blue (RGB) frequencies; a simple white light source; ultraviolet light (UV ) Source, or any other suitable radiation source. The illumination source can include, for example, 322 LEDs including 64 white LEDs, 86 red LEDs, 86 green LEDs, and 86 blue LEDs. The number of light sources at any location can vary from that shown and described herein. The RGB illumination is provided as a trio of three-color LEDs that operate at each respective frequency at each position. There can be different combinations of RGB LEDs for different illumination sources. Each type of illumination is provided by a specific card with a specific arrangement of LEDs. The base unit generates the backlight beam 306 by two cards each having a plurality of diodes arranged in three. Each trio of LEDs 404 includes red, green and blue LEDs. A total of 45 LED trios are located on each card and used to generate the backlight beam 306. Within each trio, the red, green and blue LEDs can be lit once at a time so that one color of illumination is created after the other color.

  In all the illumination examples described above, an image of the biological sample is captured from above by the image capture device 254. Note that in order to measure colony growth and other time-related effects, the image capture device 254 may capture a series of image sequence sets over a predetermined period of time. In addition, the image capture device 254 may be a video camera for certain applications where colony growth progress or the like is being measured. Petri dish movement can also be effected by moving the petri dish in and out of the imaging system 114 by a suitable conveyor belt or robotic arm.

  Image capture device 254 is adapted to take different types of images from different illumination sources. Typically, a series of images are taken for a particular application. This sequence includes illuminating the sample with a particular illumination or combination of illuminations, followed by taking a particular type of image, such as monochromatic, black and white, or RGB, with associated illumination. The next image is then taken with a different type of illumination or a combination thereof and this sequence continues until all the required images have been taken. The image capture device 254 is controlled in sequence to take the appropriate type of image.

  As mentioned above, the biological sample in the imaging system 114 can be illuminated from a plurality of different light sources that strike the sample from different directions. After the biological sample is irradiated, an image of the biological sample is taken from above. Each irradiation highlights a different aspect of the biological sample.

  The backlight illumination shows a detailed view of the petri dish details including any markings on the base of the petri dish, the edges and lid shape of the petri dish; and the placement and density of colonies in the biological sample. This illumination provides information that can isolate the colonies and determine the differences between similar colonies (eg, α and β hemolytic species), generally giving a view of the contents of the sample.

  The substantially horizontal illumination is refracted and reflected by the sample and its contents to form an image that can be used to isolate and eliminate artifacts. In addition, this image can be used to make corrections based on the absorption of irradiation by the medium, as described in more detail below. Images can also be used later to determine the coverage of petri dishes by colonies to provide an estimate of colony concentration and to determine growth or non-growth on non-opaque media.

  Annular illumination is directed at the sample and reflected or refracted to the image capture device 254 by the medium and any colonies that are formed. The purpose of the image produced by this illumination is the ability to distinguish between the color of the medium and the colony. The ability to distinguish colors is often an important tool for identifying specific microorganisms because some microorganisms have very unique colors. The overall result is the closest view to what a biologist expects to see for a particular type of microorganism, eg, color, colony appearance, etc. This is particularly important in distinguishing subtle changes in color around the media and colonies. In addition, the image generated by the annular illumination makes it possible to detect subtle changes in the color of the underside and surroundings of the bacterial colonies in the chromogenic medium.

  Side annular illumination is illumination by only one of four linearly positioned illumination sources. This provides an image with a shadow that can be used to identify contours and irregularities. Each of the illumination sources produces a different shadow effect as a result of the direction of illumination.

  Inverse annular illumination is reflected from the head unit onto the sample. The sample then reflects or refracts the illumination to the image capture device 254. Images captured in this way give contrast details of different colonies in the sample. This image can also contribute to color information. In addition, the image may provide texture information; colony appearance and color; information about the swarming limit and predetermined information about colony irregularities, such as bumps, morphology and shape.

  Reverse ring illumination creates pseudo vertical illumination that can visualize gradation changes. This gives texture and granularity information and is useful for detecting colonies with irregular surfaces, although there are no large bumps. In one implementation, a number of different images can be taken and subsequently merged using reverse annular illumination to address the possibility of image saturation.

  The vertical illumination source illuminates the biological sample from above. This illumination is reflected by biological samples and colonies to obtain an image that gives detailed contour information. This can be used to identify the roughness and colony height of a biological sample. Secondly, the colony irregularities are often very specific, so this information can be used to determine the specific type of microorganism. For example, some colonies have a dome shape, others are bumpy, others are flat, and so on.

  As described above, each of the illumination sources and directions can be used to enhance and improve different image characteristics. The described embodiments can be altered or adapted by using illumination from different illumination sources and directions without departing from the scope of the present invention.

  Furthermore, different wavelengths of illumination can be used for different applications, such as infrared and ultraviolet.

  The imaging system 114 can create an image from a combination of illumination sources to create a composite image that can enhance and enhance more than one feature of the biological sample image.

  Another factor that can have an effect on the image produced is the type of media used to grow the sample. There are many different media, which are particularly useful for identifying hemolysis ability, and CPS, which is a media adapted to identify E. coli, Proteus, and KESC, especially using urine samples; COS, which is a medium containing blood, is included.

  Different media, such as CPS and COS, are quite different in nature and color. As a result, the illumination applied to these media can produce different types of images. Thus, different illumination sources and combinations of illumination sources can be used for different media.

  As previously mentioned, the system 100 includes a processing system 108 and an analysis unit 110. The processing system 108 includes a first processing unit 116, a second processing unit 118, a calculation unit 120 and a feature extraction unit 122.

  The first processing unit 116 obtains a binary image showing a visible separation line between the contents of the Petri dish associated with the microbial colony and the contents of the Petri dish associated with the medium. A first process, which is a segmentation process applied by the imaging system 114, can be activated, applied to the original image of the Petri dish, such as a backlight or bottom annular view. Thus, the first processing unit 116 provides a first processed image that is a binary image named GrowthMask that contains visible pixels that are related only to the pixels associated with the microbial colony.

  The first processing unit 116 is a well-known integration that integrates two original images in a well-known edge detection process that finds the position of the contents of the Petri dish relative to the edge of the Petri dish and a well-known process for removing image noise. Additional processing, including processing, can also be run.

  The second processing unit 118 can operate a second process which is a dilation process applied to the GrowthMask to obtain a second processed image named Dilated (GrowthMask). . Dilated (GrowthMask) includes visible pixels that are associated only with pixels associated with microbial colonies and also with pixels associated with surrounding zones around the microbial colonies. Extended parameters are defined such that Dilated (GrowthMask) encompasses the surrounding zone. The expansion parameters are determined by the user as described below. A surrounding zone may be associated with a zone that exhibits a different color when compared to the color of microorganisms in the same image. The surrounding zone can also be associated with a zone that exhibits a discoloration of the microorganism, or a zone that exhibits new pigmentation when compared to the color of the microorganism in the same image.

  The calculation unit 120 can acquire the resultant image by operating the third process related to the difference calculation between the two images. Thus, the calculation unit 120 can calculate the original image of the Petri dish, i.e., the difference between the DiscMask and GrowthMask, to provide a third processed image named CultureMediumMask. The CultureMediumMask contains visible pixels that are related only to the pixels associated with the media.

  The calculation unit 120 can also operate a fourth process related to the difference calculation between Dilated (GrowthMask) and GrowthMask to provide a fourth processed image named HaloMask. HaloMask contains visible pixels that are relevant only to the surrounding zone.

The feature extraction unit 122 can operate a feature extraction process for determining the feature quantities of the Halo Mask and the Culture Medium Mask. The feature amount is related to the pixel values of HaloMask and CultureMediumMask. The feature extraction unit 122 calculates the difference between the average pixel value of the third processed image and the average pixel value of the fourth processed image for at least three red green blue (RGB) color channels. As described above, the term “average pixel value” refers to the arithmetic average of pixel values or the median value of pixel values.
The analysis unit 110 can operate the analysis process based on the determined feature amount in order to determine the presence of the bacteria and identify the bacteria. More specifically, the analysis unit 110 determines the value of the indicator in order to compare the value of the indicator with the determined threshold value. Depending on the result of the comparison, analysis unit 110 determines either the presence or absence of bacteria in the Petri dish.

  According to the present invention, as shown in FIG. 4, the method includes a step 400 for acquiring an original image of a Petri dish using the imaging system 114 according to specific lighting conditions. Certainly, lighting conditions can vary depending on the type of bacteria identified. The original image may be processed, for example, to detect the edges of the Petri dish.

  Next, the method obtains a first image named GrowthMask by applying a first process, which is a segmentation process, to the original image, step 402 for obtaining a first processed image. Including. The method then applies a second process, which is an extension process, to the first processed image to obtain a second image named Dilated (GrowthMask), thereby obtaining a second processed image. Step 404 is included. The method includes a further step 406 for obtaining a third processed image, applying a third process to obtain a third image named CultureMediumMask. The method also includes a step 408 for obtaining a fourth processed image, applying a fourth process to obtain a fourth image named HaloMask. The method determines a feature amount associated with the third processed image and the fourth processed image to determine a color difference between the third processed image and the fourth processed image. Step 410 is further included. The method also includes a step 412 for determining the value of the indicator by classifying the feature quantities. Next, in step 414, the value of the index is compared with a threshold value. The method includes a final step 416 that determines the presence of bacteria and identifies the bacteria in the petri dish as identified bacteria.

  In a first embodiment of the invention, the method relates to determining the presence of beta hemolytic bacteria to identify hemolysis events. The method of the first embodiment includes the following steps shown in FIG.

  In a first embodiment, HaloMask relates to a halo zone that shows a discoloration or fading zone around a microbial colony that is characteristic of hemolysis. The discoloration zone results from the color of the microorganism in the same image. The medium is blood agar.

  In a first embodiment, the method includes a step 500 for obtaining an initial image consisting of the integration of the images shown in FIGS. 7a, 7b and 7c. The initial image is acquired by using the center bottom annular view for the RGB color channel using the imaging system 114. In fact, hemolysis is less noticeable in images taken in this view, especially in the green and blue channels. FIG. 7a shows the original image for the red channel. FIG. 7b shows the original image for the green channel and FIG. 7c shows the original image for the blue channel.

  The method then includes a step 502 for determining the edge of the Petri dish using a known edge detection method and a step for removing image noise using a known Gaussian filter, where σ = 1 and k = 5 × 5.

  The method then includes step 504 for separating each color channel and integrating the image from the green channel and the image from the blue channel that have average pixel values for both images to provide an average image. Step 506 is included. Indeed, removing the image from the red channel avoids considering the halo zone as a microbial colony.

  The method further includes a segmentation process including the following steps. The segmentation process includes determining a threshold at step 508 according to the well-known Otsu method based on the integrated image of the green channel image and the blue channel image. Indeed, the Otsu method provides a threshold called threshold (Otsu) that compares the average value of the pixel values at the determined position (x, y) on the image. More specifically, the Otsu method is a threshold that minimizes the ratios associated with intra-class variance and inter-class variance, where bright pixel values are associated with the first class and dark pixel values are associated with the second class. I will provide a. In the present situation, bright pixels are associated with microbial colonies and dark pixels are associated with the culture medium.

The method then includes a step 510 for segmenting the central bottom annular image by applying the following segmentation equation (1):
here,
Is the average value of the same pixel at position (x, y) on the same image for the green and blue channels.

  Application of Otsu method provides binary images.

  The method also includes an optional step 512 for applying a morphological operation based on a well-known closing operator to the binary image to obtain a binary image called GrowthMask shown in FIG.

  The method according to the first embodiment then includes another step 514 for obtaining the second original image shown in FIG. 6 by using the backlight view in the imaging system 114. Indeed, the hemolysis phenomenon is more pronounced in the images taken in this view, the zone associated with the medium is red and the discoloration zone is transparent in white or yellow, or a composite color consisting of yellow and white The zone related to microorganisms is dark, such as black.

  The method includes a step 516 for applying a well-known expansion operation to GrowthMask, the expansion operation being based on a structured member such as a disk-shaped member. The extended calculation is performed using a known function such as a matlab function immediate and a specific extended parameter d in order to obtain an extended image corresponding to GrowthMask called Dilated (GrowthMask).

The Halo Mask shown in FIG. 9a and the Culture Medium Mask shown in FIG. 9b can be estimated from Dilated (Growth Mask) as shown below.
Here, the colony of the microorganism indicated by GrowthMask is removed from Dilated (GrowthMask) to obtain HaloMask.
Here, microbial colonies and halo zones associated with hemolysis are removed from the dish mask to yield the Culture Medium Mask.

HaloMask indicates the color around the microbial colony.
CultureMediumMask shows the original color of the medium, that is, the color that is not affected by the growth of microorganisms.

  In the next step 518, the HaloMask and the CultureMediumMask are each divided into pixel patches. 10a shows a HaloMask pixel patch, and FIG. 10b shows a CultureMediumMask pixel patch. In FIGS. 10a and 10b, each pixel patch includes 200 × 200 pixels and the size of the extension is 50 pixels. The size of the pixel patch is adapted to detect the associated median.

  The division of HaloMask and CultureMediumMask is processed with a 50% overlap ratio between two adjacent pixel patches for HaloMask and two adjacent pixel patches for CultureMediumMask. Therefore, the number of pixel patches including both halo pixels and medium pixels is optimized.

  In addition, a reference is set for the pixel patch associated with HaloMask. Pixel patches that contain regions where the halo zone contains less than 100 pixels are not considered in the next step of the method.

  In a further step 520, the HaloMask and CultureMediumMask center pixel values are computed for each pixel patch, for each channel in the RGB color space, and for the grayscale display.

  The calculation of the difference between the statistical center pixel values for HaloMask and CultureMediumMask is processed several times based on the increasing value of the expansion parameter d.

  The determination of the expansion parameter d to be selected for the next step is then processed in step 522.

The determination of the extension parameter d is based on the following formula that determines the value of the extension parameter d that maximizes the difference value between the pixel value of HaloMask and the pixel value of CultureMediumMask. The following extension equation (2) applies to each RGB color channel and grayscale display:
The increasing value of the extension parameter is selected with a width of 10 pixels for two consecutive values of d in the interval from 10 to 60 pixels.

  Step 522 provides a plurality of binary images called Dilated (GrowthMask), and thus a plurality of HaloMasks and CultureMediumMasks.

  FIGS. 11a, 12a and 13a each show an example of a HaloMask with different values of the expansion parameter d, d = 10 pixels in FIG. 11a, d = 50 pixels in FIG. 12a, and d = 60 pixels in FIG. 13a. It is.

FIGS. 11b, 12b and 13b each show an example of a Culture Medium Mask with different values of the expansion parameter d, d = 10 pixels in FIG. 11b, d = 50 pixels in FIG. 12b, and d = 60 pixels in FIG. 13b. It is.
In the present situation, the determined expansion parameter is d = 50 pixels.

Based on the determined extension parameters, the difference in median pixel values for the same channel between HaloMask and CultureMediumMask provides four difference values between the median for the RGB color channel and the median of the grayscale display. It is computed in step 524 by applying the following feature (3):

In a further step 526, the analysis unit 110 operates the step for classifying the feature patch values according to the determined rule that sets the threshold calculated using the following expected value equation (4):
Here, the Feature patch is a difference value of the central pixel value between the Halo Mask and the Culture Medium Mask on the blue channel.

  FIG. 14a shows images of Petri dishes associated with beta hemolytic bacteria, alpha hemolytic bacteria and gamma hemolytic bacteria, respectively, with a display of features at the patch level on the blue channel.

  The set of values associated with the expected value expression represents the distribution of Beta hemolysis patches and NoBeta hemolysis patches, where the Beta hemolysis patch includes a halo zone representing a hemolysis phenomenon, and the NoBeta hemolysis patch includes a halo zone representing a hemolysis phenomenon. Not included.

  In step 528, the distribution of NoBeta hemolytic patches is compared to a Gaussian distribution based on mean value μ and standard deviation σ, as shown in FIG. 14b.

FIG.
The threshold based on the following 3σ rule (5) is set to determine the normal distribution as shown below:
This means that the threshold is equal to μ + 3σ.

  In step 530, analysis unit 110 determines the presence or absence of beta hemolytic bacteria. If the Feature patches values are high positive values, these values indicate the presence of beta hemolytic bacteria.

  If the Feature patch values are negative or close to 0, these values indicate other phenomena such as the presence of alpha hemolytic or gamma hemolytic bacteria.

  In a further particular embodiment, the use of pixel patches provides a method for determining the presence of hemolysis locally, ie in a particular zone of the Petri dish image.

  According to a second embodiment of the invention relating to the determination of the presence of Proteus, the method comprises the following steps shown in FIG.

  In a second embodiment, HaloMask relates to a halo zone that shows a pigmentation zone around the microbial colony, where the color of the pigmentation zone is caused by the presence of microorganisms in the same image. The medium is associated with a chromogenic medium such as CPS.

  In a second embodiment, the method includes a step 1500 for obtaining two original views of the Petri dish by using a backlit no background view and a central bottom annular black background view. The center bottom annular black background view image is the median image of the four bottom annular side view images.

  FIG. 16a shows the original image of a Petri dish by using a backlit backgroundless view. In practice, FIG. 16a shows a dark zone associated with the presence of microbial colonies, and the pigmentation zone is associated with a zone having a brown color around the dark zone.

  The method then includes a step for determining the edge of the Petri dish with a well-known edge detection method.

  The method further includes a step 1502 for applying the specific graph-based region segmentation process disclosed in bibliographic reference [1].

  The graph-based region segmentation process includes a first region called the background associated with the media zone and a second region called the foreground associated with the rest of the petri dish and also associated with the serigraphy. Provide an image containing the zone.

  Such a segmentation process segments both clusters of microbial colonies and isolated colonies of microorganisms within the same mask, ie, the same foreground region. In addition, graph-based region segmentation is adapted to slight color changes in the media zone, as in the present situation with the presence of Proteus halo.

  The segmentation process also includes steps for detecting the serigraphy zone using well-known detection algorithms.

  The segmentation process then provides a binary image called GrowthMask, as shown in FIG. 16b.

  The method includes a further step 1504 for applying an expansion operation to GrowthMask, the expansion operation being based on a structured member, such as a disk-shaped member. The extended operation is calculated by a well-known function such as a matlab function immediate to create a Dilated (GrowthMask).

  HaloMask relates to a zone that may contain a pigmentation zone around a microbial colony, said pigmentation zone representing the presence of Proteus. CultureMediumMask shows the original color of the medium, that is, the color that is not affected by the pigmentation zone.

The HaloMask shown in FIG. 17b and the CultureMediumMask shown in FIG. 17c are estimated from Dilated (GrowthMask) and DishMask associated with the original image shown in FIG. 16a, as shown in Equations (6) and (7) below. be able to:

  The extension operation is applied several times based on the increasing value of the extension parameter d. The increasing value of the expansion parameter d is selected with a width of 10 pixels for two consecutive values of d in the interval from 10 pixels to 70 pixels. Step 1504 provides a plurality of images called Dilated (GrowthMask).

FIG. 18a shows an example of a HaloMask with an extended parameter d = 70 pixels.
FIG. 18b shows an example of a HaloMask with the extended parameter d = 10 pixels.

  In the next step, feature extraction unit 120 determines the features of HaloMask and CultureMediumMask. Therefore, the feature extraction unit 120 extracts color features.

  The method includes a step (not shown) for computing a statistical median of pixel values for RGB color space and HSV (Hue, Saturation, Lightness) space. In the HSV space, the value H is replaced with a vector (cosH, sinH) to avoid a discontinuity in the hue scale where 0 and 1 correspond to the same hue value. Thus, the HSV space includes four channels.

  Two different images are selected to calculate the statistical median. The first image is associated with a backlit no background view and the second image is associated with a central bottom annular black background view. The central bottom annular black background view image is the median image of the four bottom side view images.

The method includes a step 1506 for computing the following feature (8) for each RGB color channel and each channel in the HSV space:

  In the RGB color space, there are three RGB color channels where seven extended operations are applied to two different images. Therefore, when applying the above formula, there are a total of 42 features to be extracted.

  In HSV space, there are four channels where seven extension operations are applied to two different images. Therefore, when applying the above formula, there are a total of 56 features to be extracted.

  The total sum of the extracted features is 98.

  The analysis unit 120 then operates by applying classification steps 1508 and 1510 based on two classification algorithms used in the supervised learning process well known in the prior art.

  The first classification algorithm is related to the classification and regression tree (CART) algorithm. For example, the matlab class “ClassificationTree” is used to model the CART algorithm. Application of the first algorithm provides a first score value representing the likelihood that the Petri dish contains Proteus.

  The second classification algorithm is related to the support vector machine (SVM) algorithm and is created using the library “libSVM”. Application of the second algorithm provides a second score value representing the likelihood that the Petri dish will contain Proteus.

  Since the first score and the second score are provided by two different classification algorithms, the first score and the second score are not comparable values and are combined in the same calculation to determine the final score It is not possible.

  The classification unit 120 operates by applying a Bayes theorem to obtain a value of posterior probability, i.e., p (Halo; X), where Halo is an event representing halo.

  In the first classification algorithm, X represents the amount of features selected by the CART algorithm. The selected feature is related to the feature that is the most discriminating feature for determining the presence of Proteus among the 98 existing features.

  In the second classification algorithm, X represents the closest example distance from the decision line. The highest distance X represents a reliable result for determining the likelihood that the Petri dish contains Proteus.

The probability of the presence of a halo can be calculated with the following equation (9):
Where p (X; Halo) is the probability of observing X assuming that the Petri dish contains Halo;
-P (Halo) is the a priori probability of the presence of halo;
And p (Halo; X) is the probability of the presence of a halo given the value X.

  If the Petri dish does not contain Proteus, the probability of halos occurring is expected to be zero.

  However, it is possible that the Petri dish does not have a visual halo and the probability of having a halo also contains Proteus bacteria that is also equal to zero. Therefore, in order to facilitate the detection of Proteus, the above equation (9) is modified to avoid such a situation. Equation (10) below takes into account the hypotheses associated with the presence of visible halos for all Proteus bacteria.

The classification unit 120 then calculates the following equation (10).
Where p (X; Proteus) is the probability of X assuming that the Petri dish contains Proteus;
-And p (Proteus) is the a priori probability of the presence of Proteus.

  A set of values related to the probability p (X; Proteus) is represented in FIG. 19 as a distribution of values based on a Gaussian distribution having a mean value μ and a standard deviation σ.

  p (Proteus) is an a priori probability and is an adjustable parameter set to 1/6 by default. In fact, CPS media can contain up to 6 classes of bacteria.

p (X) is an estimate of the normalization factor calculated by the following equation (11):
p (Proteus), p (X; Proteus), p (NoProteus) and P (X; NoProteus) are known from the data set used for the supervised learning process.

  Accordingly, in a further step 1512, a first value of p (Halo; X) is calculated using a first score associated with the first classification algorithm, ie the CART algorithm.

  In a further step 1514, a second value of p (Halo; X) is calculated using a second score associated with the second classification algorithm, ie the SVM algorithm.

  Thus, the first and second values of p (Halo; X) are now related to the first and second probabilities of the same type of value, i.e. comparable values.

  In a further step 1516, the minimum of the first and second values of p (Halo; X) is obtained and the lowest minimum value between the first and second values of p (Halo; X) is selected. Is done.

  Then, in step 1516, a final probability of the presence of Proteus between 0% and 100% is obtained.

  The user decides to use a threshold such as 50% to determine the probability of the presence of Proteus in the Petri dish in Step 1520 by comparing the value of the probability of the presence of Proteus in Step 1518. can do.

  The use of two probabilities, each based on the first classification algorithm and the second classification algorithm, results in correction of errors caused independently by each classification algorithm.

  In another embodiment related to determining the presence of Proteus, the medium can be an opaque medium such as opaque CPS.

  FIG. 20 shows an original image of a Petri dish related to backlight view, including an opaque medium such as opaque CPS containing Proteus.

  21 and 22 show images of the Petri dish of FIG. 20 associated with the left bottom annular view and the top annular view, respectively.

  A further embodiment of the present invention relates to the above method relating to the detection of Proteus, wherein the culture medium is an opaque medium.

  Step 1500 and steps 1504 to 1520 associated with the above method remain the same. The step 1502 step associated with the application of the segmentation process must be adapted to take into account the presence of opaque medium. Indeed, in the presence of opaque media, the segmentation process involves the application of an adapted graph-based region segmentation algorithm, and the steps associated with detection and removal of serigraphy zones are eliminated.

  In this embodiment, the image used as the original image is based on the backlight view and the upper annular view. Other parameters such as the method of combining the original image and the threshold associated with the foreground extraction process are adapted to allow the method associated with the detection of Proteus using an opaque medium to operate in an efficient manner. It becomes.

  The following description relates to the application of the method of the invention with respect to detection of beta hemolytic bacteria and detection of Proteus.

  The method for the detection of beta hemolytic bacteria has been applied to a first data set comprising a petri dish containing a single bacterial colony and a blood agar medium such as COS and CAN medium, said petri dish being incubated for 24 hours. ing.

  The first data set contains a total of 104 petri dishes, and according to the ground truth reference, 21 petri dishes contain beta hemolytic bacteria and 83 petri dishes do not contain beta hemolytic bacteria, ie beta hemolytic Contains bacteria other than bacteria.

Table 1 below shows the results of applying this method to the first data set, where 14 out of a total of 21 Petri dishes were identified as containing beta hemolytic bacteria and a total of 83 Of the Petri dishes, 83 Petri dishes have been identified as free of beta hemolytic bacteria:

  From Table 1 it appears that when applying the method according to the present invention, seven Petri dishes containing beta hemolytic bacteria were not detected and identified. Certainly, the halo zone in the corresponding Petri dish is not clearly visible, so the color difference between the halo zone and the medium zone for the Petri dish is a contrast level sufficient to determine the presence of hemolytic bacteria. Does not bring

Table 2 below shows the same results as shown in Table 1, using 95% confidence intervals and percentages as indicators.

According to Table 2 above, the percentage of global accuracy is higher than 93%.
The method for detection of beta-hemolytic bacteria has also been applied to a second data set comprising a petri dish containing a single bacterial colony and a blood agar medium such as COS and CAN medium, which is incubated for 24 hours. Has been.

  The second data set contains a total of 70 petri dishes, according to the ground truth reference, 24 petri dishes contain beta hemolytic bacteria and 46 petri dishes do not contain beta hemolytic bacteria, ie beta hemolysis. Includes bacteria other than sex bacteria.

Table 3 below shows the results of applying this method to the second data set, out of a total of 241 Petri dishes, 23 Petri dishes were identified as containing beta hemolytic bacteria, Of the 46 Petri dishes, 43 Petri dishes have been identified as free of beta hemolytic bacteria:

  From Table 3, when applying the method according to the present invention, it appears that one Petri dish containing beta hemolytic bacteria was not detected and identified. In fact, the Petri dish contained only one microbial colony and the segmentation process did not provide the correct Culture Medium Mask and the correct Halo Mask.

  From Table 3, it also appears that three false detections of beta hemolytic bacteria occurred. These false detections can occur by the presence of a halo zone around the colony of gamma hemolytic bacteria, by the presence of a serigraphy zone, or by the presence of the edge of a Petri dish. However, such false detections can be read out in the method steps associated with the threshold determination by the Otsu method.

Table 4 below shows the same results as shown in Table 1 by using percentages as indicators with a 95% confidence interval:

According to Table 4 above, the percentage of global accuracy is higher than 94%.
The results shown in Tables 1, 2, 3 and 4 show a high level of accuracy of 95% for the method of the invention for the detection of beta hemolytic bacteria.

  The method for detection of Proteus is applied to the first data set for a Petri dish containing a single bacterial colony and a mixture of bacteria from two different classes of bacteria and a translucent chromogenic medium, said Petri dish being 24 hours Incubated.

  The first data set includes Proteus having a halo zone and Proteus having no halo zone.

  As mentioned above, the method according to the present invention for detecting Proteus bacteria is supervised to separate the feature data into two different classes that are the presence or absence of a halo zone. Classifying features by using a model algorithm. The same data set is used for each supervised model algorithm.

The analysis of the results is based on Table 5 below, which is a confusion matrix that determines the level of confusion for the supervised model used.

  In the current situation, the available ground truth is the presence of Proteus or the absence of Proteus, whereas the method provides predictions related to the presence or absence of the halo zone.

Considering the learning model for classification and regression trees, the results shown in Table 6 below are obtained:

Table 6 shows that the accuracy level is 99% which is a very good level.
The sensitivity level is 78% which is also a good level. As described above, the data set includes a Petri dish containing Proteus without a halo zone. A visual inspection of 22% of the Petri dishes listed in Table 6 indicates that the majority of the Petri dishes contain Proteus bacteria that do not have a halo zone. Moreover, it seems that the boundary of the colony is erroneously detected as a halo zone by the supervised model algorithm. However, the use of statistical center pixel values overcomes this problem.

Considering the learning model for SVM, the results shown in Table 7 below are obtained:

  The percentage of Petri dishes with halo zones and no Proteus bacteria is different from the corresponding percentages read in Table 6 for classification and regression trees.

  When considering opaque media, the results for the detection of Proteus are quite similar to those read with translucent media. In addition, since the halo zone is more visible in Petri dishes containing opaque media, this method has less false negative detection, ie Proteus is classified as no halo.

  From disclosure: CART keeps its unique node with the same feature selected and a unique extension size equal to 10 pixels.

Table 8 below shows an example of a confusion matrix for applying the SVM algorithm:

  As explained above with respect to the Proteus detection embodiment, the reliability index is a percentage calculated by the Bayesian theorem by using the posterior probabilities provided by the two supervised model algorithms CART and SVM. is there. The value of the reliability index belongs to the interval [0; 100], 0% means that there is no halo zone, and 100% means that there is a halo zone in the Petri dish.

as mentioned above,
Where p (X; Proteus) is the probability of X assuming that the distance to the margin is X and that the Petri dish contains Proteus;
And p (Proteus) is the a priori probability of the presence of Proteus.

FIG. 23 shows the distribution of probabilities obtained for both the CART and SVM supervised model algorithms, and the probabilities for Petri dishes with and without Proteus in both situations with a semi-transparent medium. Indicates.
Circles indicate detection of false positive results shown in Tables 6 and 7.

  FIG. 23 also shows the distribution of final probabilities representing the merged value Pf of the lowest probabilities Ptree and Psvm for both the supervised model algorithms CART and SVM.

The 50% threshold is selected as follows:

In Table 9 below, the reliability index results are evaluated.

  Table 9 above shows that the accuracy level is higher than 99%, which is a very good level. Indeed, the merging of the results provided by the two supervised model algorithms avoids false positive detection.

  The sensitivity level is 71%, which is also a good level because some Petri dishes contain Proteus bacteria that do not contain a halo zone.

  FIG. 24 shows the distribution of probabilities obtained for both CART and SVM of the supervised model algorithm, for both situations, including Petri dishes containing Proteus and opaque media, and Proteus containing opaque media Shows the probability of no petri dish.

FIG. 24 also shows the distribution of final probabilities representing the merged value Pf of the lowest probabilities Ptree and Psvm for both CART and SVM of the supervised model algorithm.

In another situation where the Petri dish was incubated for 18 hours, the results shown in Table 11 below were obtained for the translucent medium.

The corresponding results for the opaque medium are shown in Table 12 below.

  As shown in Tables 11 and 12 above, the results obtained over the 18 hour and 24 hour periods are stable for Petri dishes containing translucent or opaque media.

Bibliographic reference
[1] Pedro F. Felzenszwalb, Daniel P. Huttenlocher, “Efficient Graph-Based Image Segmentation”; International Journal of Computer Vision, September 2004, Vol. 59, No. 2, pp. 167-181

Claims (18)

A method for determining the presence of at least one identified microorganism in a Petri dish comprising one or more colonies of microorganisms and a medium, wherein the medium is present with one or more colonies of microorganisms In a method adapted to allow said at least one identified microorganism to grow under suitable growth conditions,
-Obtaining at least one initial image of the Petri dish, wherein the first initial image comprises one or more visible pixels, each pixel being associated with a pixel value;
-Obtaining a first processed image of the Petri dish by applying a first process to at least one initial image, wherein the visible pixels of the first processed image are one or more colonies of microorganisms; Steps associated only with pixels associated with
-Obtaining a plurality of second processed images of the Petri dish by applying a second process to the first processed image, wherein the visible pixels of the second processed image are one or more of the microorganisms; Relating only to pixels associated with a plurality of colonies and surrounding zones around said one or more colonies of microorganisms;
-Obtaining a plurality of third processed images of the Petri dish by calculating a difference between at least one initial image and a plurality of second processed images, the plurality of third processing The completed image includes visible pixels associated only with pixels associated with the culture medium;
-Obtaining a plurality of fourth processed images of the Petri dish by calculating a difference between the plurality of second processed images and the first processed image, the plurality of fourth processed images The visible pixels of the processed image are related only to the pixels associated with the surrounding zone;
By calculating a difference value between an average pixel value of a plurality of third processed images and an average pixel value of a plurality of fourth processed images for at least each color channel of a red green blue (RGB) color channel; Determining features associated with the plurality of third and fourth processed images;
-Determining the value of the indicator of the presence of at least one identified microorganism in the surrounding zone in the Petri dish by classifying the determined features;
-Comparing the value of the indicator with a threshold;
-Determining the presence of at least one identified microorganism in the surrounding zone in the Petri dish, depending on the result of the comparison.   The method of claim 1, wherein the first process is a segmentation process.   The method of claim 2, wherein the segmentation process includes determining a threshold and comparing a pixel value to the threshold.   The method of claim 2, wherein the segmentation process is a graph-based region segmentation process.   The method of claim 1, wherein the second process is an expansion process associated with an expansion criterion having a plurality of defined values.   The method of claim 1, wherein determining the feature amount comprises dividing the plurality of third processed images and the fourth processed image into pixel patches.   The method of claim 6, further comprising: for each pixel patch, calculating a maximum pixel value difference between pixel values of the third processed image and the fourth processed image.   The method according to claim 7, wherein the step of determining the feature amount includes calculating a maximum difference of pixel values for the grayscale image.   The step of determining the feature amount comprises calculating a center pixel value of the plurality of first processed images and a center pixel value of the plurality of first processed images relative to the HSV space. Method.   The method of claim 1, wherein the step of classifying the feature quantity comprises comparing the feature quantity distribution with a Gaussian distribution.   The method of claim 1, wherein classifying features comprises applying a classification and regression tree algorithm and a support vector algorithm to provide corresponding first and second scores.   The method of claim 11, wherein classifying features further comprises applying a Bayes theorem to the first and second scores.   The method of claim 12, wherein the step of classifying the feature further comprises obtaining an integration possibility of the presence of surrounding zones.   The method of claim 1, wherein comparing the value of the indicator with a threshold includes determining the threshold based on an average of the Gaussian distribution and a standard deviation σ.   15. Any of claims 1, 2, 3, 5, 6, 7, 8, 10, or 14 wherein determining the presence of at least the identified microorganism comprises determining the presence of a bacterium that is a beta hemolytic bacterium. The method according to claim 1.   14. The method of any one of claims 1, 4, 9, 11, 12, or 13, wherein the step of determining the presence of at least the identified microorganism comprises determining the presence of a bacterium that is Proteus. A system for determining the presence of at least one identified microorganism in a Petri dish comprising one or more colonies of microorganisms and a medium, wherein the medium is present with one or more colonies of microorganisms An imaging system adapted to allow the at least one identified microorganism to grow under suitable growth conditions, and a processing system for acquiring at least one initial image of the Petri dish (108), the processing system (108) comprises:
-A first processing unit for obtaining a first processed image of at least one initial image of the petri dish, wherein the visible pixels of the first processed image are associated with one or more colonies of the microorganism; A first processing unit (116) associated only with the selected pixels;
-A second processing unit for obtaining a plurality of second processed images of the first processed image of the petri dish, wherein the visible pixels of the second processed image are one or more colonies of the microorganism; And a second processing unit (118) relating only to pixels associated with the surrounding zone around the one or more colonies of microorganisms;
-Obtaining a plurality of third processed images of the Petri dish by calculating a difference between the second initial image and the second processed image; and the second processed image and the first processed image A calculation unit for obtaining a plurality of fourth processed images of a Petri dish by calculating the difference between the processed images, wherein the third processed image is only for the pixels associated with the medium. A computing unit (120) associated with the visible pixels of the fourth processed image relating only to pixels associated with the surrounding zone;
A plurality of third processes by calculating, for at least each color channel of the RGB color channel, a difference value between the average pixel value of the third processed image and the average pixel value of the fourth processed image; A feature extraction unit (122) for determining a feature quantity associated with the completed image and the fourth processed image;
Determining the value of the indicator of the presence of at least one identified microorganism in the surrounding zone in the Petri dish and comparing the value of the indicator with a threshold; and depending on the result of the comparison, the surrounding zone in the Petri dish An analysis unit (110) for determining the presence of at least one identified microorganism in the
A system (100) comprising:   A computer program product comprising instructions that, when executed, cause a programmable data processing device to perform the steps of the method according to any of the preceding claims.